GPU Performance while Encoding H.265 Video

[cspells]

Hello,

My team is wondering if the GPU performance for AI inference will be affected by simultaneously encoding video on the Jetson Orin? We found the following architecture diagram which shows that the data flows directly from the CSI x4 input to the Video Ingest and then to the NV Encoder.

It seems that the GPU performance will not be degraded by simultaneously encoding four video streams because the encoder is on separate hardware, but we wanted to verify this.

Also, we are using SDI camera data, and we are looking at bridge hardware to bring the SDI stream in through PCIe or with CSI x4. Is the CSI x4 much faster than bringing in camera data from the PCIe? Will the PCIe input achieve the same encoding speed as with CSI x4?

Any help is much appreciated.

Thanks.

[DaneLLL]

Hi,
PCIe/CSI is the interface of capturing frame data and you may check how many sources you have in your use-case and choose the interface. Generally the frame data is captured through v4l2 and we run the gstreamer pipeline for encoding:

nvv4l2camerasrc ! 'video/x-raw(memory:NVMM),format=UYVY' ! nvvidconv ! 'video/x-raw(memory:NVMM),format=NV12' ! nvv4l2h264enc ! h264parse ! matroskamux ! filesink

For encoding speed, please refer to module data sheet:
https://developer.nvidia.com/jetson-agx-xavier-series-datasheet

[cspells]

Thanks for the command; I will use that during benchtop testing. I am looking more for results from testing or specs about how much the GPU performance for AI inference will be affected by simultaneous video encoding. Does anyone have results from tests that they have run or more information about this?

Thanks.

[DaneLLL]

Hi,
We don’t have the data. Since NVENC and GPU are independent hardware engines, it shall not impact each other.